The goal of this study is to explore the advantages of representing natural images with the cortical Layer-4 processing, which is the first step in visual information processing performed by the cerebral cortex of the brain. A cortical module, a macrocolumn, receives input from a small visual field and its Layer 4 performs a nonlinear transform of this input to generate its pluripotent representation. In this study, we design some tests on such image windows in order to explore the differences and advantages of Layer-4 representation over the pixel representation. These tests measure how much the neighborhood is preserved in the feature space and how much discriminability remains between spatially related (neighboring/shifted) image windows. The accuracies of the representations are measured using Support Vector Machines (SVM) and K-Nearest Neighbor (K-NN) algorithms as the classification methods.